Presentation to the 2nd International Workshop on Flame Chemistry, preceding the 35th International Symposium on Combustion, in San Francisco, CA, in August 2014.
Describes recent progress in two projects related to our Reaction Mechanism Generator software.
1. .edu/comocheng
Reaction Mechanism Generator:
Toward High-Throughput Transition State Calculations,
and Interpreting Existing Kinetic Models
Richard H. West
4 August 2014
1
r.west@neu.edu richardhwest rwest
Presented at the 2nd International Workshop on Flame Chemistry in San Francisco
2. Automatic reaction mechanism generation
needs methods to:
1. Represent molecules
(and identify duplicates)
The common theme to the two projects in the talk is Reaction Mechanism Generator software.
First, a brief introduction to it.
3. 2. Create reactions
(and then new species)
Automatic reaction mechanism generation
needs methods to:
CH3
+
4. Automatic reaction mechanism generation
needs methods to:
3. Choose which reactions to include
(and which to leave out)
You’ll notice a lot of slide designs are top-heavy.
This is because the room layout meant most people couldn’t see the bottom of the screen!
7. Reaction families propose all possible
reactions with given species
•Template for recognizing reactive sites
•Recipe for changing the bonding at the site
•Rules for estimating the rate
There are ~40 reaction families such as as Hydrogen abstraction,
unimolecular homolysis, radical addition to a double bond…
8. All possible reactions are found, then core is
expanded by following the fastest pathways.
A
B
C
D
E
F
G
H
A
B
C
D
E
F
9. F
G
H
Model can be started from “seed mechanism”
(if you have this in RMG format)
Getting it in “RMG format” is the hard part, and will be addressed in the second half of the talk.
10. Rate estimates are based on the local
structure of the reacting sites.
• Hydrogen abstraction: XH + Y. → X. + YH
• Rate depends on X and Y.
O
H
O
11. Rate estimation rules are organized in a tree
Part of the tree for X
The most generic expression goes at the top of the tree.
You use the most precise node that you can, then fall up until you find data.
13. New “RMG-Py” designed to be
more extensible and developer-friendly.
• See poster on Wednesday
We presented two posters describing this work at the 35th International Symposium on Combustion,
occurring the week following this presentation.
14. New “RMG-Py” designed to be
more extensible and developer-friendly.
• See poster on Wednesday
We presented two posters describing this work at the 35th International Symposium on Combustion,
occurring the week following this presentation.
15. New “RMG-Py” designed to be
more extensible and developer-friendly.
• See poster on Wednesday
We presented two posters describing this work at the 35th International Symposium on Combustion,
occurring the week following this presentation.
16. .edu/comocheng
Toward High-Throughput
Transition State Calculations
Pierre L. Bhoorasingh
14
Acknowledgment is made to the Donors of
the American Chemical Society Petroleum
Research Fund for support of this research.
Now that you understand RMG, here’s the first of the two projects.
17. Ideal tree: lots of data
Ideally we climb the tree, find the matching node for our reaction,
and pluck the leaf containing the rate expression.
21. Heat release (red) and progress variable (purple) of a turbulent lean methane-air Bunsen flame simulated using S3D. DOE
At my first Combustion Symposium (in 2006) I was impressed by
the computer resources used by the CFD modelers
22. Heat release (red) and progress variable (purple) of a turbulent lean methane-air Bunsen flame simulated using S3D. DOE
16,000,000
CPU-hours
At my first Combustion Symposium (in 2006) I was impressed by
the computer resources used by the CFD modelers
23. Assoc. Lib Director, ORNL Computing
This is one of the large federal computers they use.
Imagine what we could do if combustion chemists had 16M CPU hours!…
26. Automatic TS searches remain an important
energy research goal
“An accurate description of the often
intricate mechanisms of large-molecule
reactions requires a characterization of all
relevant transition states... Development
of automatic means to search for
chemically relevant configurations is the
computational-kinetics equivalent of
improved electronic structure methods.”
- Basic Research Needs for Clean and
Efficient Combustion of 21st Century
Transportation Fuels.
US Dept of Energy (2006)
I skipped this slide, because the previous talk had made the motivation clear.
27. Automatic TS searches remain an important
energy research goal
!
“...transformation from by-hand
calculations of single reactions
to automated calculations of
millions of reactions would be a
game-changer for the field of
chemistry, and would be a good
‘Grand Challenge’ target...”
- Combustion Energy Frontier
Research Center (2010)
First Annual Conference of the
Combustion Energy Frontier Research
Center (CEFRC)
September 23-24, 2010
Princeton
I skipped this slide too.
28. Can you predict TS geometries
from molecular groups alone?
!
(this would be great)
29. Length of bond being broken,
at TS for Hydrogen abstraction
Can you predict TS geometries
from molecular groups alone?
Radical
Molecule
30. Length of bond being broken,
at TS for Hydrogen abstraction
Can you predict TS geometries
from molecular groups alone?
in Å with M06-2X/6-31+G(d,p)
!"!#$ !"!%% !"!&' !"!($
!")() !")'& !"*+$ !"*!#
!")(' !")$% !"*%%
!")'+ !"*+& !"*&) !"*&$
31. Can you predict TS geometries
from molecular groups alone?
!"!#$ !"!%% !"!&' !"!($
!")() !")'& !"*+$ !"*!#
!")(' !")$% !"*%%
!")'+ !"*+& !"*&) !"*&$
!"#$# !"#$%!"!#$ !"#$%
in Å with M06-2X/6-31+G(d,p)
Guess which number goes in the blank?
32. You can predict TS geometries
from molecular groups alone!
!"!#$ !"!%% !"!&' !"!($
!")() !")'& !"*+$ !"*!#
!")(' !")$% !"*%%
!")'+ !"*+& !"*&) !"*&$
!"#$%
in Å with M06-2X/6-31+G(d,p)
So you can predict distances! But…
(1) I gave you 15 numbers and you gave be one; and
(2) you only gave me a distance, not a 3D geometry.
33. Add the two numbers to get any distance
within 1 pm
in Å with M06-2X/6-31+G(d,p)
0.607 0.622 0.658 0.666
0.525
0.657
0.680
0.691
0.607 0.622 0.658 0.666
0.525
0.657
0.680
0.691
First challenge: getting more data out than you put in.
Now, I give you 8 numbers and you can give me 16!
Arrange these groups in a tree as mentioned before, and job done.
34. Use distance geometry to position atoms
⇌RMG
Molecule
Connectivity
3D
Structure
Second challenge: getting from a distance to a 3D structure.
We use the distance-geometry algorithms in RDKit.
35. Use distance geometry to position atoms
⇌RMG
Molecule
Connectivity
Atoms List
AtomsList
Upper limits
Lower limits
Generate bounds
matrix
Embed
in 3D
Second challenge: getting from a distance to a 3D structure.
We use the distance-geometry algorithms in RDKit.
36. Use distance geometry to position atoms
⇌RMG
Molecule
Connectivity
Atoms List
AtomsList
Upper limits
Lower limits
Generate bounds
matrix
Atoms List
AtomsList
Embed in 3D
Edit
bounds matrix
Second challenge: getting from a distance to a 3D structure.
We use the distance-geometry algorithms in RDKit.
37. Estimate geometry directly via
group additive distance estimates
• Database arranged in tree structure as for kinetics
• Trained on successfully optimized transition states
• Direct guess much faster than double ended search
• Success depends on training data
Generate
Bounds
Matrix
Edit Bounds
Matrix!
close to TS
Embed
Matrix in
3D
Reaction
from RMG
Optimize TS
geometry
Generate
Bounds
Matrix
Edit Bounds
Matrix!
close to TS
Embed
Matrix in
3D
Double-
ended
SearchReactants
Products
IRC!
Calculation
Generate
Bounds
Matrix
Edit Bounds
Matrix!
for TS
Embed
Matrix in
3D
38. Test the group-additive method on
H-abstraction reactions in Diisopropyl Ketone
A coordinated investigation of the combustion chemistry of diisopropyl
ketone, a prototype for biofuels produced by endophytic fungiJoshua W. Allen a
, Adam M. Scheer b
, Connie W. Gao a
, Shamel S. Merchant a
, Subith S. Vasu c
, Oliver Welz b
,
John D. Savee b
, David L. Osborn b
, Changyoul Lee d
, Stijn Vranckx d
, Zhandong Wang e
, Fei Qi e
,
Ravi X. Fernandes d,f
, William H. Green a,⇑
, Masood Z. Hadi g,1
, Craig A. Taatjes b,⇑
a
Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
b
Combustion Research Facility, Sandia National Laboratories Livermore, CA 94551, USA
c
Mechanical and Aerospace Engineering, University of Central Florida, Orlando, FL 32708, USA
d
Physico-Chemical Fundamentals of Combustion, RWTH Aachen University, 52056 Aachen, Germany
e
National Synchrotron Radiation Laboratory, University of Science and Technology of China, Hefei, Anhui 230029, PR China
f
Physikalisch-Technische Bundesanstalt, Bundesallee 100, D38116 Braunschweig, Germany
g
Biomass Conversion Technologies, Sandia National Laboratories, Livermore, CA 94551, USA
a r t i c l e i n f o
Article history:
Received 25 July 2013
Received in revised form 18 September 2013Accepted 18 October 2013
Available online 20 November 2013
Keywords:
Diisopropyl ketone
Automatic mechanism generation
Ignition delay
Pyrolysis
Combustion
Detailed kinetics modeling
a b s t r a c t
Several classes of endophytic fungi have been recently identified that convert cellulosic biomass to a
range of ketones and other oxygenated molecules, which are potentially viable as biofuels, but whose
oxidation chemistry is not yet well understood. In this work, we present a predictive kinetics model
describing the pyrolysis and oxidation of diisopropyl ketone (DIPK) that was generated automatically
using the Reaction Mechanism Generator (RMG) software package. The model predictions are evaluated
against three experiments that cover a range of temperatures, pressures, and oxygen concentrations:
(1) Synchrotron photoionization mass spectrometry (PIMS) measurements of pyrolysis in the range
800–1340 K at 30 Torr and 760 Torr; (2) Synchrotron PIMS measurements of laser photolytic Cl-initiated
oxidation from 550 K to 700 K at 8 Torr; and (3) Rapid-compression machine measurements of ignition
delay between 591 K and 720 K near 10 bar. Improvements made to the model parameters, particularly
in the areas of hydrogen abstraction from the initial DIPK molecule and low-temperature peroxy
chemistry, are discussed. Our ability to automatically generate this model and systematically improve
its parameters without fitting to the experimental results demonstrates the usefulness of the predictive
chemical kinetics paradigm.
Ó 2013 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
1. Introduction
Liquid petroleum-based fuels account for more than 30% of
world energy consumption, the majority occurring in the transpor-
tation sector, where liquid fuels dominate because of their high en-
ergy density and ease of storage and distribution [1]. However,
insecurity of oil supply and negative impacts of climate change
are motivating increased development of alternative liquid fuels
[2]. Lignocellulosic biofuels are a promising renewable alternative
to conventional petroleum fuels that could lower greenhouse gas
emissions. Due to the recalcitrance of lignocellulosic materials,
new means for breaking down biomass into fermentable sugars
are a central area of biofuels research [3]. However, the products
that are most efficiently produced from biomass may not have
well-characterized combustion properties. Furthermore, advanced
clean, efficient combustion strategies, particularly those that rely
on compression ignition, are often very sensitive to fuel chemistry
[4,5]. To address these issues comprehensively, coordinated efforts
towards biofuel-engine co-development are required.Figure 1 depicts the Sandia collaborative biofuel development
strategy. Synthetic biologists working in close collaboration with
combustion researchers develop fundamental mechanisms for
the production and combustion of potential biofuels. Ignition and
engine trials then provide feasibility tests for fuels and mixtures
and yield recommendations for the bioengineering scale-up of spe-
cific metabolic pathways. This coupling of fundamental and ap-
plied combustion chemistry and synthetic biology is a strategy to
identify and investigate the most promising fuel compounds
through mutual feedback. Moreover, the development of combus-
tion models will provide the predictive capability needed for even-
tual efficient utilization of the new biofuel stream. A similar
0010-2180/$ - see front matter Ó 2013 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.combustflame.2013.10.019
⇑ Corresponding authors.
E-mail addresses: whgreen@mit.edu (W.H. Green), cataatj@sandia.gov (C.A.
Taatjes).
1
Present address: NASA Ames Research Center, Moffett Field, CA 94035, USA.
Combustion and Flame 161 (2014) 711–724
Contents lists available at ScienceDirect
Combustion and Flame
journal homepage: www.elsevier.com/locate/combustflame
Allen, J. W. et al. Combust. Flame 161, 711–724 (2014).
These data were hot off the press.
We tried finding every H-abstraction reaction in this large published kinetic model.
39. Parity plots of Estimated against Optimized
X-H distances at the transition state
0.7
1.6
0.7
1.6
EstimatedDistanceXH(Å)
Optimized Distance XH (Å)
Parity
We’ll see how well we predict the distances as we re-train the groups on more and more data.
40. Trained on 44 reactions,
estimates not so great…
0.7
1.6
0.7
1.6
EstimatedDistanceXH(Å)
Optimized Distance XH (Å)
Training Data
44
Parity
41. Trained on 148 reactions,
estimates improve…
0.7
1.6
0.7
1.6
EstimatedDistanceXH(Å)
Optimized Distance XH (Å)
Training Data
44
148
Parity
42. Trained on 230 reactions,
estimates improve…
0.7
1.6
0.7
1.6
EstimatedDistanceXH(Å)
Optimized Distance XH (Å)
Training Data
44
148
230
Parity
43. Trained on 767 reactions,
estimates are very good.
0.7
1.6
0.7
1.6
EstimatedDistanceXH(Å)
Optimized Distance XH (Å)
Training Data
44
148
230
767
Parity
46. Transforming data into knowledge—ProcessInformatics for combustion chemistry
Michael Frenklach *Department of Mechanical Engineering, University of California, and Environmental Energy Technologies
Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
Abstract
The present frontier of combustion chemistry is the development of predictive reaction models,
namely, chemical kinetics models capable of accurate numerical predictions with quantifiable uncertain-
ties. While the usual factors like deficient knowledge of reaction pathways and insufficient accuracy of
individual measurements and/or theoretical calculations impede progress, the key obstacle is the incon-
sistency of accumulating data and proliferating reaction mechanisms. Process Informatics introduces a
new paradigm. It relies on three major components: proper organization of scientific data, availability
of scientific tools for analysis and processing of these data, and engagement of the entire scientific com-
munity in the data collection and analysis. The proper infrastructure will enable a new form of scien-
tific method by considering the entire content of information available, assessing and assuring mutual
scientific consistency of the data, rigorously assessing data uncertainty, identifying problems with the
available data, evaluating model predictability, suggesting new experimental and theoretical work with
the highest possible impact, reaching community consensus, and merging the assembled data into new
knowledge.
Ó 2006 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
Keywords: Kinetics; Modeling; Optimization; Consistency; Informatics; PrIMe
1. Introduction
1.1. Progress in combustion depends on reliablereaction kinetics
Research is motivated by the human desire toexplain phenomena surrounding us and the passionfor new discoveries. Driven by individual satisfac-
tion such activities aim at improving the qualityof human experience. One can generalize the stimu-lus for research as developing the ability of makingpredictions. Indeed, predicting properties of a mate-
rial before it is synthesized, predicting performance
of a device before it is built, or predicting the mag-nitude of a natural disaster ahead of time all have
obvious benefits to society.In the area of combustion, we have all the aboveelements: the fascination with fire from ancienttimes, its broad application to numerous aspectsof everyday life, and the definite need for reliable
predictions (see Fig. 1). In recent decades the com-
bustion research was largely driven by the desiredincrease in energy efficiency, reduction in pollutantformation, and material synthesis. The need fortying the research to practical applications was fur-ther emphasized in the Hottel address of the 28th
Symposium by Professor Glassman [1].
1540-7489/$ - see front matter Ó 2006 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.proci.2006.08.121
*
Fax: +1 510 643 5599.E-mail address: myf@me.berkeley.edu
Proceedings of the Combustion Institute 31 (2007) 125–140
www.elsevier.com/locate/proci
Proceedings
of the
Combustion
Institute
Again, we go back to my first Combustion Symposium in 2006,
where I was introduced to the idea of Process Informatics / Data Collaboration.
47. “the key obstacle is the inconsistency of accumulating
data and proliferating reaction mechanisms.
Process Informatics introduces a new paradigm. It relies
on three major components:
• proper organization of scientific data,
• availability of scientific tools for analysis and
processing of these data,
• and engagement of the entire scientific community
in the data collection and analysis.”
M. Frenklach, Proc. Combust. Inst. 31 (2006)
I assumed the entire scientific community would go home and properly organize their scientific data…
48. Model Complexity is Increasing
2-methylalkanes/Sarathy
n-Heptane/Mehl
n-Octane/Ji
Heptamethylnonane/Westbrook
Isobutene/Dias
Ethanol/Leplat
1-butanol/Grana
Acetaldehyde/Leplat
Ethylbenzene/Therrien
Methylcrotonate/Bennadji
Methyloleate/Naik
Furan/Yasunaga
Methylhexanoate/Westbrook
Methydecanoate/Herbinet
H2
/KlippensteinH2
/ConnaireH2
/Marinov
GRI-Mech
Methane/Leeds
Ethane
C2
/Karbach
Ethylene/Egolfopoulos
C3
/Appel
USC-MechMarinov
Propane/Qin
n-Butane
Neopentane/Wang
n-Heptane/Curran
n-Heptane/Babushok
10
100
1000
10000
1995 2000 2005 2010
Numberofspeciesinmodel
Year of publication
So is everything published since 2006 in PrIMe format?
No. It’s mostly in CHEMKIN format.
(This chart is very incomplete random sampling)
49. And again, here is a photo of a large federal computer. This time at NASA
50. This is the computer model Gordon and McBride used when they devised the NASA polynomial format
for thermochemistry (now used in CHEMKIN). It was designed to fit on 80-column punch-cards.
51. This is the computer model Gordon and McBride used when they devised the NASA polynomial format
for thermochemistry (now used in CHEMKIN). It was designed to fit on 80-column punch-cards.
52. APPENDIX D
THERMO DATA (FORMAT AND LISTING)
The order and format of the input data cards in this appendix are given in the follow-
ing table"
Card
order
(a)
(Final
card)
Contents
Format Card
column
THERMO
Temperature ranges for 2 sets of coefficients:
lowest T. common T, and highest T
Species name
Date
Atomic symbols and formula
Phase of species (S, L, or G for solid, liquid,
or gas, respectively)
Temperature range
Integer t
Coefficients ai(i = 1 to 5) in equations (90) to (92)
(for upper temperature interval)
Integer 2
Coefficients in equations (90) to (92) (a6, a T for
upper temperature interval, and a 1, a2, and a 3
for lower)
Integer 3
Coefficients in equations (90) to (92) (a4, a5, a6,
a, i for lower temperature interval)
Integer 4
Repeat cards numbered 1 to 4 in cc 80 for each
species
END (Indicates end of thermodynamic data)
3A4
3F10.3
3A4
2A3
4(A2, F3. O)
A1
2F10.3
I15
5(E15.8)
15
4(E15.8)
I20
3A4
1 to6
1 to 30
1 to 12
19 to 24
25 to 44
45
46 to 65
80
1 to75
80
1 to75
80
1 to 60
8O
ito3
aGaseous species and condensed species with only one condensed phase can be in
any order. However. the sets for two or more condensed phases of the same
species must be adjacent. If there are more than two condensed phases of a
species, their sets must be either in increasing or decreasing order according
to their temperature intervals.
168
ORIGINAL PAGE IS
OF. POOR QUALITY
nq
This is the computer model Gordon and McBride used when they devised the NASA polynomial format
for thermochemistry (now used in CHEMKIN). It was designed to fit on 80-column punch-cards.
53. NASA (Chemkin) format is very dense –
not much room for species identifiers
Chemical Formula Parameters for H(T), S(T)Species Name
54. Space constraints led to “creative”
naming schemes
MVOX
IIC4H7Q2-T
C3H5-A C3H5-SC6H101OOH5-4
TC4H8O2H-I
C4H8O1-3
C3KET21
CH3COCH2O2H
Notice hydroperoxypropan-2-one has two different names, often in the same mechanism file!
55. Space constraints led to “creative”
naming schemes
• C3KET21 is generated from alkyl peroxy radical
isomerization pathway
• CH3COCH2O2H is generated from low-temperature
oxidation of acetone
MVOX
IIC4H7Q2-T
C3H5-A C3H5-SC6H101OOH5-4
TC4H8O2H-I
C4H8O1-3
C3KET21
CH3COCH2O2H
Notice hydroperoxypropan-2-one has two different names, often in the same mechanism file!
56. Human-Computer team can
identify species more quickly
• Our new tool uses RMG to generate reactions to
compare with the target model
• A human reviews the evidence and confirms matches.
Generate reactions,
check for equivalents,
propose matches
Review evidence,
confirm matches.
proposed
matches
proposed
matches
Human Computer
confirmed
matches
confirmed
matches
57. Identify small molecules first
• Identify species with only one possible structure
• CO2
• H2O
• C3H8
• Then species with "borrowed" thermochemistry
• Then boot-strap based on how these react
58. Identifying ‘sc3h5co’ from its reactions
7
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
The first 6 reactions are all H-abstractions giving 4 possibilities for sc3h5co.
The 7th reaction is beta-scission, also giving 4 possibilities.
59. Identifying ‘sc3h5co’ from its reactions
7
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
The first 6 reactions are all H-abstractions giving 4 possibilities for sc3h5co.
The 7th reaction is beta-scission, also giving 4 possibilities.
60. Identifying ‘sc3h5co’ from its reactions
7
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
The first 6 reactions are all H-abstractions giving 4 possibilities for sc3h5co.
The 7th reaction is beta-scission, also giving 4 possibilities.
61. Identifying ‘sc3h5o’ from its reactions
R1-R6
R7
7
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
By considering all seven reactions, we can deduce the most likely isomer.
63. Interact with a Web UI
• Computers make mistakes, and humans can see
patterns
• All matches must be approved by an operator using a
web user interface
64. Two undergraduates this summer imported models from recent volumes of the journals
Combustion & Flame and Proceedings of the Combustion Institute. More details on our poster.
65. Two undergraduates this summer imported models from recent volumes of the journals
Combustion & Flame and Proceedings of the Combustion Institute. More details on our poster.
66. Identified 8,299 species from 58 models
• Malformed chemkin files
• Incorrect glossary entries
• 100 kJ/mol disagreements in enthalpies of formation
• 30 orders of magnitude disagreements in rates
• Only about 10% pressure-dependent
We frequently hear we can get 1 kJ/mol errors in energy and 30% in rate constants,
yet our supplementary material differ by 100 kJ/mol and 10
30
respectively!
These discrepancies are usually undetected.
67. Some recent developments in RMG
• Developer-friendly "RMG-Py" being released
• New features (Nitrogen!)
• Web tools
• Group-Additive Transition State Estimates
• Fast estimates of TS geometries
• Gets better at guessing the more it guesses
• Mechanism Importer tool
• Facilitates identification of species in chemkin files
• Started to collect and curate data
Final Summary!
68. .edu/comocheng
Richard H. West
4 August 2014
54
r.west@neu.edu richardhwest rwest
And of course, http://www.slideshare.net/richardhwest